Performance
venomx keeps a full terminal emulator per pane in the daemon and another in
each attached client, and still aims to be at least as fast as the
multiplexers it replaces. test/muxbench.py measures that by driving each
multiplexer’s real client on a PTY, the way a person would.
Results
200×50 pane, 16 MiB flood, 5 trials, medians, venomx built with ReleaseFast. AMD Ryzen 7 9800X3D, Linux 7.3; tmux 3.8, zellij 0.45.1, tuios 0.12.1, each with its default config.
Text with heavy styling (256-colour foreground plus bold or dim on every word):
| Metric | Better | venomx | tmux | zellij | tuios |
|---|---|---|---|---|---|
| Flood throughput (MiB/s) | higher | 92.4 | 25.1 | 59.6 | 73.7 |
| Flood CPU (s) | lower | 0.28 | 0.47 | 0.32 | 0.53 |
| Startup (ms) | lower | 14.8 | 23.8 | 77.4 | 69.3 |
| Echo latency p50 / p90 (ms) | lower | 0.37 / 0.57 | 0.14 / 0.19 | 11.37 / 11.45 | 2.14 / 2.60 |
| Idle memory, PSS (MiB) | lower | 11.5 | 4.3 | 94.7 | 80.0 |
| Memory after flood (MiB) | lower | 19.7 | 6.4 | 162.1 | 121.2 |
Plain ASCII:
| Metric | Better | venomx | tmux | zellij | tuios |
|---|---|---|---|---|---|
| Flood throughput (MiB/s) | higher | 177.9 | 77.1 | 37.2 | 80.5 |
| Flood CPU (s) | lower | 0.15 | 0.21 | 0.49 | 0.46 |
| Startup (ms) | lower | 15.7 | 21.8 | 77.4 | 68.8 |
| Echo latency p50 / p90 (ms) | lower | 0.39 / 0.58 | 0.14 / 0.19 | 11.36 / 11.48 | 2.14 / 2.57 |
| Idle memory, PSS (MiB) | lower | 11.5 | 4.3 | 94.9 | 79.9 |
| Memory after flood (MiB) | lower | 19.2 | 5.6 | 136.2 | 104.6 |
The defaults are not equal on every count: tmux keeps 2000 lines of history, venomx 10000.
What is measured
- Startup: from launch until the pane’s first text shows.
- Echo latency: from a keystroke to its echo on screen.
- Flood: throughput and CPU time while a pane prints 16 MiB.
- Memory: PSS summed over each multiplexer’s client and server processes, idle and after the flood.
The harness plays the host terminal: it reads all output and answers the usual queries (device attributes, cursor position, colours, size), but draws nothing, so the numbers measure the multiplexer and not a terminal emulator. Each trial starts in fresh HOME and XDG directories, and the first trial is a warmup.
Run it yourself
zig build bench-compare -Doptimize=ReleaseFast # every installed multiplexer
zig build bench-compare -Doptimize=ReleaseFast -- --data sgr --report sgr.json
python3 test/muxbench.py --compare old.json new.json # deltas; >10% regressions marked !Multiplexers that are not installed are skipped. --mux, --tmux,
--zellij and --tuios choose which run and which binaries they use.